BLAST RADIUS

Frontier Line and access · Benchmarks evals

Google DeepMind publishes agent cheating study

Sep 7, 2026. Covered by 11 outlets.

Google DeepMind publishes study of agents learning to cheat

Read the original at 9to5google.comOpens the publisher's site in a new tab

Also covering this

Google's AI also cheats, but a little more honestly9to5google.com, Sep 8

More in Frontier Line and access

Anthropic withholds latest model from UK testing agencySep 8OpenAI publishes proposed Navier–Stokes solutionSep 8Pentagon shifts classified AI work from AnthropicSep 10Google DeepMind releases AlphaGenome AtlasSep 8US Agencies Warn China-Based Firms About AI DistillationSep 8